Two-phase flows are preponderant in industrial components. The present work deals with external two-phase flows across tube banks commonly found in heat exchangers, boilers and steam generators. The flows are generally highly complex and remain theoretically intractable in most cases. The two-phase flow patterns provide a convenient albeit qualitative means for describing and classifying two phase flows. The flow patterns are also closely correlated to fluid-structure interaction dynamics and thus provide a practically useful basis for the study of two-phase flow-induced vibrations.
For internal two-phase flows, maps by Taitel et al. (1980) and others have led to detailed and well defined maps. For transverse flows in tube bundles, there is significantly less agreement on the flow patterns and governing parameters. The complexity of flow in tube arrays is an obvious challenge. A second difficulty is the definition of distinct flow patterns and the identification of parameters uniquely identifying the flow patterns.
The present work addresses the problem of two-phase flow pattern identification in tube arrays. Flow measurements using optical as well as flow visualization via high-speed videos and photography have been conducted. To identify the flow patterns, an artificial intelligence machine learning approach was taken. Pattern classification was achieved by designing a support vector machine (SVM) classifier. The SVM achieves quantitative and non-subjective classification by mapping the flow patterns in a high dimensional mathematical space in which the different flow patterns have unique characteristics.
Details of the flow measurement, parameter definition and SVM design are presented in the paper. Flow patterns identified using the SVM are presented and compared with previously identified flow patterns.